Sign In
insight - Exemplar-Free Class Incremental Learning
Diffusion-Based Class Incremental Learning: Addressing Catastrophic Forgetting with Exemplar-Free Methods
The author proposes a novel exemplar-free CIL method that addresses catastrophic forgetting by bridging domain gaps and balancing stability and plasticity. By utilizing multi-distribution matching diffusion models, selective synthetic image augmentation, and multi-domain adaptation, the method achieves state-of-the-art performance in various settings.
DiffClass: Diffusion-Based Class Incremental Learning
提案された手法は、多様な分布にマッチングする拡散モデルを使用し、安定性と可塑性のバランスを取ることで、先行研究を凌駕する結果を達成した。
DiffClass: Diffusion-Based Class Incremental Learning
새로운 클래스를 배우면서 이전 지식을 잊지 않는 방법
DiffClass: Diffusion-Based Class Incremental Learning
Neuer Ansatz zur Bewältigung von katastrophalem Vergessen und Ungleichgewicht zwischen Stabilität und Plastizität in exemplarfreiem CIL.
Products | Resources
© 2024 by Linnk AI